Hi!

On Tue, Sep 06, 2022 at 04:19:01PM +0200, Florian Weimer wrote:
> * наб via Libc-alpha:
> 
> > This is a trivial patch, largely duplicating the extant ASCII code
> >
> > There are two user-facing changes:
> >   * nl_langinfo(CODESET) is "POSIX" instead of "ANSI_X3.4-1968"
> >   * mbrtowc() and friends return b if b <= 0x7F else <UDF00>+b
> >
> > Since Issue 7 TC 2/Issue 8, the C/POSIX locale, effectively:
> >   (a) is 1-byte, stateless, and contains 256 characters
> >   (b) which collate in byte order
> >   (c) the first 128 characters are equivalent to ASCII (like previous)
> > cf. https://www.austingroupbugs.net/view.php?id=663 for a summary of
> > changes to the standard;
> > in short, this means that mbrtowc() must never fail and must return
> >   b if b <= 0x7F else ab+c for all bytes b
> >   where c is some constant >=0x80
> >     and a is a positive integer constant
> >
> > By strategically picking c=<UDF00> we land at the tail-end of the
> > Unicode Low Surrogate Area at DC00-DFFF, described as
> >   > Isolated surrogate code points have no interpretation;
> >   > consequently, no character code charts or names lists
> >   > are provided for this range.
> > and match musl
> 
> We don't match Python and its surrogateescape encoding (PEP 838).
404?

> It
> maps invalid bytes in the 0x80…0xff range to U+DC80…U+DCFF.
(The same as musl.)

> It may make
> more sense to align with that.
With a=1 and c=<UDF00>, assuming it's as you say, we very much do?

$ printf '\x80\xff' | output/elf/ld.so --library-path output/ output/iconv/iconv_prog -fPOSIX -tUCS4 | hd
00000000  00 00 df 80 00 00 df ff                           |........|
00000008

> Anyway, regarding mechanics, we'll need a new localedata/charmaps/POSIX
> charmap, I think.  This charmap then can be tested against the gconv
> converter.
Hm, the problem with that is tst-tables -> tst-table -> tst-table-from
(and -to) convert by constructing a UTF-8 sequence. The problem with
this approach is that glibc rejects unpaired surrogates.

The output for tst-table-from UTF-8 is:
  ...
  0xED9FBE        0xD7FE
  0xED9FBF        0xD7FF
  0xEE8080        0xE000
  0xEE8081        0xE001
  ...
i.e. there's a gap for the surrogates; and, indeed, the charmap reads
  <UD7FB>     /xed/x9f/xbb HANGUL JONGSEONG PHIEUPH-THIEUTH
  %<UD800>     /xed/xa0/x80 <Non Private Use High Surrogate, First>
  %<UDB7F>     /xed/xad/xbf <Non Private Use High Surrogate, Last>
  %<UDB80>     /xed/xae/x80 <Private Use High Surrogate, First>
  %<UDBFF>     /xed/xaf/xbf <Private Use High Surrogate, Last>
  %<UDC00>     /xed/xb0/x80 <Low Surrogate, First>
  %<UDFFF>     /xed/xbf/xbf <Low Surrogate, Last>
  <UE000>..<UE03F> /xee/x80/x80 <Private Use>
with the surrogate range commented-out;
this dates back to the inclusion of UTF-8 generator scripts in 2015
(4a4839c94a4c93ffc0d5b95c69a08b02a57007f2), these exclusions are
deliberate (grep for surrog in localedata/unicode-gen/utf8_gen.py).

Given this limitation, expanding the charmap to
ANSI_X3.4-1968 + <UDF80>..<UDFFF> doesn't actually test much:
having them as separate codepoints will always fail tests,
and dot-notation lines are ignored when generating the comparison
tables, so this particular type of test just proves that POSIX is the
same as ANSI_X3.4-1968 for the first 128 characters.

There's already an exhaustive iconv_prog-based testsuite
(cf. additions to iconv/tst-iconv_prog.sh), though.

> You should put the new converters into a separate file (not
> iconv/gconv_simple.c), then the s390x version will use that
> automatically.
Oh, of course! Moved to iconv/gconv_posix.c.

> > diff --git a/localedata/locales/POSIX b/localedata/locales/POSIX
> > index 7ec7f1c577..fc34a6abc1 100644
> > --- a/localedata/locales/POSIX
> > +++ b/localedata/locales/POSIX
> > @@ -97,6 +97,20 @@ END LC_CTYPE
> >  LC_COLLATE
> >  % This is the POSIX Locale definition for the LC_COLLATE category.
> 
> Isn't this just the C locale?
Yes, C is defined to be POSIX.

> We don't have a separate file for that.
Yes, we very obviously do, seeing as this patch edits it?
Nothing consumes it AFAICT, but.

> > diff --git a/wcsmbs/wcsmbsload.c b/wcsmbs/wcsmbsload.c
> > index 0f0f55f9ed..f87099bcf5 100644
> > --- a/wcsmbs/wcsmbsload.c
> > +++ b/wcsmbs/wcsmbsload.c
> > @@ -33,10 +33,10 @@ static const struct __gconv_step to_wc =
> >    .__shlib_handle = NULL,
> >    .__modname = NULL,
> >    .__counter = INT_MAX,
> > -  .__from_name = (char *) "ANSI_X3.4-1968//TRANSLIT",
> > +  .__from_name = (char *) "POSIX",
> >    .__to_name = (char *) "INTERNAL",
> > -  .__fct = __gconv_transform_ascii_internal,
> > -  .__btowc_fct = __gconv_btwoc_ascii,
> > +  .__fct = __gconv_transform_posix_internal,
> > +  .__btowc_fct = __gconv_btwoc_posix,
> >    .__init_fct = NULL,
> >    .__end_fct = NULL,
> >    .__min_needed_from = 1,
> > @@ -53,8 +53,8 @@ static const struct __gconv_step to_mb =
> >    .__modname = NULL,
> >    .__counter = INT_MAX,
> >    .__from_name = (char *) "INTERNAL",
> > -  .__to_name = (char *) "ANSI_X3.4-1968//TRANSLIT",
> > -  .__fct = __gconv_transform_internal_ascii,
> > +  .__to_name = (char *) "POSIX",
> > +  .__fct = __gconv_transform_internal_posix,
> >    .__btowc_fct = NULL,
> >    .__init_fct = NULL,
> >    .__end_fct = NULL,
> 
> This makes the comment on __wcsmbs_gconv_fcts_c in the same file
> obsolete.

Comment fixed.

> Thanks,
> Florian

New patchset in followup.

Best,
наб